Has there been any progress on this?
I can get 2x 4090s working via NCCL_P2P_DISABLE=1. But moving to 3x 4090s barely improves throughput over 2x 4090s.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Parallel training with 4 cards 4090 cannot be performed on AMD 5975WX， stuck at the beginning | 14 | 5965 | February 20, 2023 | |
| P2P not working for P600s? | 7 | 1816 | April 5, 2018 | |
| P2P issue using two RTX 5090 GPUs | 12 | 1419 | March 16, 2025 | |
| NVidia driver 520.61.05 / Cuda 11.8 / RTX 3090 = black display and superslow modesets | 21 | 24810 | December 6, 2022 | |
| Low P2P GPU bandwidth performance between GeForce GPUs | 20 | 985 | October 9, 2024 | |
| Using GTX 590 cards for CUDA SLI cards under CUDA? | 37 | 14242 | April 2, 2012 | |
| P2P access not enabled, is this a software or a hardware issue? | 7 | 9703 | November 10, 2015 | |
| why "all CUDA-capable devices are busy or unavailable" ? | 34 | 64452 | April 20, 2011 | |
| Keep getting "GPU has fallen off the bus" with 3090 cards on Gigabyte MZ32-AR1 Rev 3.0 motherboard | 18 | 329 | June 10, 2025 | |
| GPU Utilization Drops after Consecutive Executions | 28 | 5728 | October 2, 2013 |